Save VRAM, Compute, and Cost — Small Models Can Do This Too!